Compare IQR, Z-score, and modified Z-score outputs. Inspect flagged values, summary metrics, and positions fast. Improve dataset quality before training, scoring, monitoring, and reporting.
| Index | Example Value | Note |
|---|---|---|
| 1 | 12 | Typical series point |
| 2 | 13 | Typical series point |
| 3 | 14 | Typical series point |
| 4 | 12 | Typical series point |
| 5 | 15 | Typical series point |
| 6 | 16 | Higher but still near cluster |
| 7 | 14 | Typical series point |
| 8 | 13 | Typical series point |
| 9 | 97 | Strong candidate outlier |
| 10 | 12 | Typical series point |
| 11 | 11 | Low normal point |
| 12 | 15 | Typical series point |
IQR method: IQR = Q3 - Q1. Lower Fence = Q1 - (Threshold × IQR). Upper Fence = Q3 + (Threshold × IQR). Values beyond the chosen fence are flagged.
Z-Score method: Z = (x - mean) / sample standard deviation. A value is flagged when its score passes the selected threshold.
Modified Z-Score method: Modified Z = 0.6745 × (x - median) / MAD. MAD is the median absolute deviation from the series median.
Outlier Detection Series Calculator helps analysts review unusual values inside ordered numeric data. It supports machine learning preparation, anomaly screening, and data quality checks. You can test a sensor stream, model feature list, experiment batch, or validation series before training.
Outliers can shift averages and distort variance. They can also break scaling, reduce model stability, and create false alerts. A clear detection workflow helps teams inspect rare events without guessing. It also improves trust in dashboards, forecasts, and predictive pipelines.
This calculator includes IQR, Z-Score, and Modified Z-Score detection. IQR works well for skewed data and uses quartiles. Z-Score compares each value to the mean and standard deviation. Modified Z-Score uses the median and MAD, so it is more robust when extreme values already exist.
Feature engineering often starts with clean numeric series. This tool helps you inspect values before scaling, normalization, clustering, regression, and classification. It is also helpful for drift review, threshold tuning, monitoring, and rapid exploratory analysis.
The result area reports summary metrics and flagged observations. You can see the index, value, score, and outlier status for each point. That makes auditing easier. You can compare methods, adjust thresholds, and confirm whether a value is truly unusual or simply rare.
Use IQR when you want a simple spread-based rule. Use Z-Score when data is fairly symmetric and standard deviation is meaningful. Use Modified Z-Score when you want better resistance to extreme values. Comparing methods can reveal whether a point is consistently abnormal.
Paste a numeric series, choose a method, set a threshold, and review the output. Then export the table for documentation or team review. This creates a repeatable process for cleaning datasets, validating inputs, and improving downstream model performance.
Series-based review is useful because order often matters. A sudden spike may signal sensor failure, data entry error, fraud, or a genuine event. This calculator does not delete values automatically. Instead, it supports informed decisions. That balance is important in responsible machine learning, where context should guide treatment during model governance.
Start with IQR for a quick and simple review. Use Modified Z-Score when you expect strong extremes. Use Z-Score when your series is roughly symmetric and standard deviation is informative.
A common IQR threshold is 1.5. A common Z-Score threshold is 3. A common Modified Z-Score threshold is 3.5. Adjust thresholds when your data has special business or scientific rules.
No. It identifies candidates only. You should review domain context before removing, capping, or transforming any value. Some extreme points are errors, but others are valid rare events.
Yes. It works on numeric sequences and ordered observations. Still, it does not model seasonality or trend. For advanced time series anomalies, use this as a screening step before deeper analysis.
Each method measures unusual behavior differently. IQR uses quartiles, Z-Score uses mean and spread, and Modified Z-Score uses median and MAD. That changes sensitivity, especially with skewed or noisy data.
If all values are nearly identical, Z-Score or Modified Z-Score may produce zero scores. That means the series lacks enough spread for that method to separate unusual points well.
Sometimes yes. High-only checks help with spikes, fraud, or overuse. Low-only checks help with drops, underperformance, or missing-like values. Both-side checks are best for general cleaning.
CSV helps with spreadsheet review, versioning, and further modeling. PDF helps with sharing, signoff, and audit records. Exporting keeps your outlier review process repeatable and easier to document.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.